MCAN-KAN: A Novel Multi-Scale Attention End-to-End Network for Fault Diagnosis of Industrial Robots

Abstract

Industrial robots play a significant role in the current industry. However, as industrial robots operate in complex industrial environments, crucial defect characteristics are tough to discover using conventional deep learning algorithms. To extract key fault features and achieve industrial robots' state monitoring, the study provides a multiscale convolutional attention network based on KAN (MCAN-KAN). Firstly, to extract rich features, a multi-scale feature fusion layer was designed. Secondly, to enhance the multi-scale fusion features, a dual-scale feature enhancement layer was designed. To fully utilize the capabilities of the two stages, a multi-stage attention characteristic interaction layer was created. Finally, KAN is used as the classifier to further improve the diagnostic performance of MCAN-KAN. The reliability of the MCAN-KAN approach has been verified utilizing two datasets. Experiments show that MCAN-KAN is superior to the existing intelligent fault diagnosis algorithms.